Rob Park at SFI

"Logic and Intent: Shaping Today's Financial Markets"

  • Started program trading with a spread algo between Deere and Caterpillar, under the assumption that fundamental drivers were similar and spreads will revert to mean. 
    • In executing this algo, felt orders were being copied by someone else.
  • Today, 70% of total US volume is algos.
  • How do algos introduce risks?
    • Problems occur when you can’t predict.
  • The algo ecosystem: the number of possibilities grow exponentially when algos interact with other algos.
    • 1 runaway algo problem. Example-on Amazon there was a $1 million book. Someone raised the price in marketplaces of another ever so slightly and that triggered a cascade where this book ended up listed for $20 million (the story of how this happened is fascinating and told here)
    • 2 Flash crash – unpredictable interaction of algos
  • What is an algorithm? It is a sequence of logic statements. All algos are created by humans. They do what people intend them to do. Intent=important. Humans are driven by incentives, algorithms are driven by human intent.
    • The technologist needs to understand the human goal, or else risk is introduced into the system.
  • IEX introduced a 350 microsecond delay on an order reaching the exchange.
  • The broker’s dilemma: brokers were matching orders between buyers and sellers, so brokers created dark pools. Broker A gets the buy, Broker B gets the sell, what’s the incentive for Broker A to trade with B?
  • In today’s market there are 11 exchanges, 40+ dark pools (IEX right now is a dark pool, but will try to become an exchange eventually).
  • Exchange dilemma: exchanges facilitate issuers with investors. Exchanges are supposed to be neutral to all participants, but now are for-profit companies who build services for specific customers. This is not the intended purpose of exchanges, and biases these exchanges towards one kind of participant (HFTs) over another.
  • There have been three generations of market algos so far:
    • 1 automatic traders flow, algos execute upon traders’ ideas, helping these traders focus on “their work” as opposed to execution
    • 2 gaming automatic trader-based algos. These algos took advantage of transparent inefficiencies in the first generations functionality.
    • 3 counteract generation 2. A trader who wants to buy size needs to game level two algos in order to hide intent and execute efficiently.
  • Participants send orders, but they don’t arrive at the actual exchange at the same time.
  • At the micro level, markets are deterministic (opposite of physics).
  • Latency arb—in a distributed system, race conditions matter. HFT aims to exploit the race. Exchanges need to know where the market is before pricing a transaction.  Introducing the 350 microsecond delay through a fishing-line like fiber. In doing so, assume the order is not fast. And then figure out where the market is.
  • Resistance to IEX so far has come from 2nd generation algo programmers.